In a single afternoon, a journalist built three Android applications using Google's AI Studio, a tool that converts natural language prompts into fully functional mobile apps. The experience showcased both the remarkable speed of modern AI coding and the significant limitations that still exist in creating polished, reliable software.

The Experiment: How AI Studio Works

Google's AI Studio, powered by the Gemini model, allows users to describe an app idea in plain English and then automatically generates the corresponding code, designs, and even installs the app onto a connected Android device. The process begins when a user types a prompt into the AI Studio web interface. The AI then attempts to autocomplete the idea, suggesting additional features, and then proceeds to write the entire codebase without further human intervention.

In this test, the journalist started with a simple request: "Make me a Doom-like text adventure game called MOOD, where MOOD stands for Modern Online Oratory Dungeon." Gemini immediately suggested additional elements such as procedural generation, turn-based combat, hidden secrets, and a progression system. The journalist accepted most suggestions, and within 20 minutes, the AI had produced five design mockups and a fully installable application.

The Three Apps Created

MOOD: A Doom-Inspired Text Adventure

The first app, MOOD, was a text-based adventure game set in a dungeon. The AI wrote a narrative involving an AI called "The Core Orator" that turns internet outrage into corporate profits. Players could defeat it by attacking, merging, or entering a backdoor password. However, the game had several shortcomings: the writing was poor, the dungeon contained only 11 rooms, and the player could win by mashing the attack button. Hidden secrets were not actually hidden; they were presented as glowing buttons. The AI even exposed the backdoor password at the moment it was needed, removing any challenge.

Calorie Counter App

The second app was a calorie counter that attempted to estimate calories in food items. The AI initially relied on the paid Gemini API for calorie lookup, but when that failed, it fell back to a hardcoded database. This led to significant inaccuracies—for example, a 16-ounce boba milk tea was estimated at only 190 calories because the AI matched "milk" and assumed skim milk. The journalist reported that a serving of Taiwanese popcorn chicken was listed at 140 calories, likely half the real value. Bug fixes were possible when the AI correctly identified the issue, but the underlying reasoning was often flawed.

Super Peach Rescue: A Nintendo Clone

The third app was a side-scrolling platformer inspired by Super Mario, starring Princess Peach. The AI generated the game with recognizable Mario elements—power-ups, pipes, enemies—but the result was barely functional. The game crashed whenever Peach touched a power-up block, and the second pipe was impossible to traverse because the character could not jump high enough. The AI could not fix these bugs during the test session.

Speed vs. Quality

The most impressive aspect of AI Studio was the sheer speed. From prompt to installed app, the process took minutes. Bug fixes could be applied nearly as quickly: the journalist reported that when a missing button caused a game-breaking bug in MOOD, he described the problem to the AI, and within seconds a new version was installed with the issue resolved. This iterative process felt close to the "vibe coding" ideal where users can rapidly prototype ideas without traditional programming skills.

However, quality suffered considerably. All three apps had fundamental design flaws, poor user experience, and inaccuracies. The calorie counter's data was unreliable, the games had broken mechanics, and the AI-generated narratives lacked depth. Moreover, the free tier of AI Studio imposed a daily limit, which interrupted the workflow just as the journalist began iterating to improve the apps.

Context and Implications

Google announced AI Studio's app-building capabilities during Google I/O 2026, positioning it as a tool that democratizes software creation. The company demonstrated a Doom-like game on stage, and the journalist's colleague had already built a personal workout tracker that they found genuinely useful. The promise is that non-programmers can create custom software to solve niche problems—like a custom calorie tracker—without hiring developers.

Yet the current state of AI coding is reminiscent of early generative AI text and image tools: impressive demos, but production-ready output remains elusive. The AI can generate code that compiles and runs, but it struggles with consistency, edge cases, and deep understanding of user intent. The journalist noted that the AI's default choices often reflected generic patterns rather than the specific needs of the user.

Limitations Encountered

Beyond bugs and inaccuracies, the daily usage cap highlighted a monetization strategy that may frustrate early adopters. After a handful of iterations, the journalist was told he had reached his limit and would need to pay or wait. This upsell felt premature, as the apps were not yet functional enough to justify a subscription.

Another limitation is the AI's tendency to over-suggest features that the user did not ask for, leading to feature creep and unnecessary complexity. In the case of MOOD, the AI added branching dialogues and multiple endings, but the implementation was shallow. Similarly, the calorie counter's reliance on an API call for every item made it slow and expensive, while the fallback database was incomplete.

Legal and ethical concerns also arise. The Super Peach Rescue app closely mimics Nintendo's copyrighted characters and gameplay, raising questions about whether AI Studio should allow such outputs. Google's terms of service may prohibit copyright infringement, but the AI generated the clone without objection.

The journalist concluded that while he was happy his own games were bad—because he prefers supporting human game developers—he could see the appeal of using AI for personal utility apps. The calorie counter, if refined, could be a genuinely useful tool that no commercial provider would build for an individual.

Overall, Google AI Studio represents a significant step forward in making app development accessible. It reduces the barrier from idea to prototype to near zero. But the gap between a prototype and a reliable, usable application remains wide. As AI models improve and incorporate better reasoning, the quality gap may narrow. For now, the technology serves as a powerful brainstorming and rapid prototyping tool, but not yet as a replacement for skilled developers.

Source: The Verge News